Longitudinal modeling of multimorbidity trajectories using large language models

Multimorbidity, the co-occurrence of two or more chronic conditions within an individual, is a major and escalating global health challenge, complicating treatment regimens, straining healthcare resources, and worsening patient outcomes. The complex interplay of shared genetic predispositions, biological pathways, and socioeconomic factors underpins its development, but clinical and research efforts have largely focused on managing diseases in isolation. Understanding multimorbidity trajectories--the accumulation and interaction of chronic diseases over time--is essential to improving preventive strategies and optimizing personalized care. Here, we introduce ForeSITE (Forecasting Susceptibility to Illness with Transformer Embeddings), a novel, transformer-based framework that harnesses advanced machine learning to predict multimorbidity progression. By analyzing longitudinal data from 480,000 participants in the UK Biobank, ForeSITE identifies distinct patterns in the co-occurrence and timing of diseases. Our temporal disease network provides insights into how certain diseases might share common genetic, environmental, or socioeconomic factors, offering more specific guidance for earlier detection and more effective disease management.